Performance Analysis of Ensemble Learning Models Comparing Bagging and Boosting Techniques for Early Preeclampsia Risk Detection in Pregnant Women Prediction

Abstract

Preeclampsia is a major pregnancy complication that substantially contributes to maternal morbidity and mortality worldwide, making early identification of risk factors essential for effective prevention and timely clinical intervention. This study evaluates the performance of ensemble learning models by comparing bagging and boosting techniques to develop an accurate early prediction system for preeclampsia risk using clinical medical record data. The research follows the Cross-Industry Standard Process for Data Mining (CRISP-DM) framework, encompassing data understanding, data preparation, modeling, evaluation, and interpretation. The dataset was obtained from RSUD Inche Abdoel Moeis Samarinda and underwent preprocessing procedures, including data cleaning, transformation, feature encoding, normalization, and Synthetic Minority Over-sampling Technique (SMOTE) to address class imbalance. Six ensemble learning algorithms were evaluated, consisting of Random Forest, Extra Trees, and Rotation Forest as bagging methods, and XGBoost, LightGBM, and CatBoost as boosting methods. Model performance was assessed using accuracy, precision, recall, and weighted F1-score. The experimental results demonstrate that Random Forest achieved the highest predictive performance, with an accuracy of 0.92, precision of 0.93, recall of 0.92, and weighted F1-score of 0.91, indicating superior robustness and generalization capability. Extra Trees achieved comparable accuracy (0.92) but exhibited lower prediction stability across evaluation metrics. Among the boosting algorithms, LightGBM and CatBoost each obtained an accuracy of 0.89, while XGBoost achieved 0.88. Rotation Forest recorded the lowest accuracy (0.62), suggesting limited suitability for this clinical dataset. These findings indicate that bagging-based ensemble methods, particularly Random Forest, outperform boosting techniques for imbalanced clinical data and provide strong empirical support for developing reliable Clinical Decision Support Systems (CDSS) for early preeclampsia screening and risk assessment in healthcare settings